3 * Copyright © 2014-2016 Canonical, Inc
4 * Author: Serge Hallyn <serge.hallyn@ubuntu.com>
6 * See COPYING file for details.
9 #define FUSE_USE_VERSION 26
24 #include <linux/sched.h>
25 #include <sys/param.h>
26 #include <sys/socket.h>
27 #include <sys/mount.h>
28 #include <sys/epoll.h>
33 #include "config.h" // for VERSION
38 LXC_TYPE_PROC_MEMINFO
,
39 LXC_TYPE_PROC_CPUINFO
,
42 LXC_TYPE_PROC_DISKSTATS
,
51 char *buf
; // unused as of yet
53 int size
; //actual data size
57 /* reserve buffer size, for cpuall in /proc/stat */
58 #define BUF_RESERVE_SIZE 256
61 * A table caching which pid is init for a pid namespace.
62 * When looking up which pid is init for $qpid, we first
63 * 1. Stat /proc/$qpid/ns/pid.
64 * 2. Check whether the ino_t is in our store.
65 * a. if not, fork a child in qpid's ns to send us
66 * ucred.pid = 1, and read the initpid. Cache
67 * initpid and creation time for /proc/initpid
68 * in a new store entry.
69 * b. if so, verify that /proc/initpid still matches
70 * what we have saved. If not, clear the store
71 * entry and go back to a. If so, return the
74 struct pidns_init_store
{
75 ino_t ino
; // inode number for /proc/$pid/ns/pid
76 pid_t initpid
; // the pid of nit in that ns
77 long int ctime
; // the time at which /proc/$initpid was created
78 struct pidns_init_store
*next
;
82 /* lol - look at how they are allocated in the kernel */
83 #define PIDNS_HASH_SIZE 4096
84 #define HASH(x) ((x) % PIDNS_HASH_SIZE)
86 static struct pidns_init_store
*pidns_hash_table
[PIDNS_HASH_SIZE
];
87 static pthread_mutex_t pidns_store_mutex
= PTHREAD_MUTEX_INITIALIZER
;
88 static void lock_mutex(pthread_mutex_t
*l
)
92 if ((ret
= pthread_mutex_lock(l
)) != 0) {
93 fprintf(stderr
, "pthread_mutex_lock returned:%d %s\n", ret
, strerror(ret
));
98 static void unlock_mutex(pthread_mutex_t
*l
)
102 if ((ret
= pthread_mutex_unlock(l
)) != 0) {
103 fprintf(stderr
, "pthread_mutex_unlock returned:%d %s\n", ret
, strerror(ret
));
108 static void store_lock(void)
110 lock_mutex(&pidns_store_mutex
);
113 static void store_unlock(void)
115 unlock_mutex(&pidns_store_mutex
);
118 /* Must be called under store_lock */
119 static bool initpid_still_valid(struct pidns_init_store
*e
, struct stat
*nsfdsb
)
124 snprintf(fnam
, 100, "/proc/%d", e
->initpid
);
125 if (stat(fnam
, &initsb
) < 0)
128 fprintf(stderr
, "comparing ctime %ld %ld for pid %d\n",
129 e
->ctime
, initsb
.st_ctime
, e
->initpid
);
131 if (e
->ctime
!= initsb
.st_ctime
)
136 /* Must be called under store_lock */
137 static void remove_initpid(struct pidns_init_store
*e
)
139 struct pidns_init_store
*tmp
;
143 fprintf(stderr
, "remove_initpid: removing entry for %d\n", e
->initpid
);
146 if (pidns_hash_table
[h
] == e
) {
147 pidns_hash_table
[h
] = e
->next
;
152 tmp
= pidns_hash_table
[h
];
154 if (tmp
->next
== e
) {
164 /* Must be called under store_lock */
165 static void prune_initpid_store(void)
167 static long int last_prune
= 0;
168 struct pidns_init_store
*e
, *prev
, *delme
;
169 long int now
, threshold
;
173 last_prune
= time(NULL
);
177 if (now
< last_prune
+ PURGE_SECS
)
180 fprintf(stderr
, "pruning\n");
183 threshold
= now
- 2 * PURGE_SECS
;
185 for (i
= 0; i
< PIDNS_HASH_SIZE
; i
++) {
186 for (prev
= NULL
, e
= pidns_hash_table
[i
]; e
; ) {
187 if (e
->lastcheck
< threshold
) {
189 fprintf(stderr
, "Removing cached entry for %d\n", e
->initpid
);
193 prev
->next
= e
->next
;
195 pidns_hash_table
[i
] = e
->next
;
206 /* Must be called under store_lock */
207 static void save_initpid(struct stat
*sb
, pid_t pid
)
209 struct pidns_init_store
*e
;
215 fprintf(stderr
, "save_initpid: adding entry for %d\n", pid
);
217 snprintf(fpath
, 100, "/proc/%d", pid
);
218 if (stat(fpath
, &procsb
) < 0)
221 e
= malloc(sizeof(*e
));
225 e
->ctime
= procsb
.st_ctime
;
227 e
->next
= pidns_hash_table
[h
];
228 e
->lastcheck
= time(NULL
);
229 pidns_hash_table
[h
] = e
;
233 * Given the stat(2) info for a nsfd pid inode, lookup the init_pid_store
234 * entry for the inode number and creation time. Verify that the init pid
235 * is still valid. If not, remove it. Return the entry if valid, NULL
237 * Must be called under store_lock
239 static struct pidns_init_store
*lookup_verify_initpid(struct stat
*sb
)
241 int h
= HASH(sb
->st_ino
);
242 struct pidns_init_store
*e
= pidns_hash_table
[h
];
245 if (e
->ino
== sb
->st_ino
) {
246 if (initpid_still_valid(e
, sb
)) {
247 e
->lastcheck
= time(NULL
);
259 static int is_dir(const char *path
)
262 int ret
= stat(path
, &statbuf
);
263 if (ret
== 0 && S_ISDIR(statbuf
.st_mode
))
268 static char *must_copy_string(const char *str
)
280 static inline void drop_trailing_newlines(char *s
)
284 for (l
=strlen(s
); l
>0 && s
[l
-1] == '\n'; l
--)
288 #define BATCH_SIZE 50
289 static void dorealloc(char **mem
, size_t oldlen
, size_t newlen
)
291 int newbatches
= (newlen
/ BATCH_SIZE
) + 1;
292 int oldbatches
= (oldlen
/ BATCH_SIZE
) + 1;
294 if (!*mem
|| newbatches
> oldbatches
) {
297 tmp
= realloc(*mem
, newbatches
* BATCH_SIZE
);
302 static void append_line(char **contents
, size_t *len
, char *line
, ssize_t linelen
)
304 size_t newlen
= *len
+ linelen
;
305 dorealloc(contents
, *len
, newlen
+ 1);
306 memcpy(*contents
+ *len
, line
, linelen
+1);
310 static char *slurp_file(const char *from
, int fd
)
313 char *contents
= NULL
;
314 FILE *f
= fdopen(fd
, "r");
315 size_t len
= 0, fulllen
= 0;
321 while ((linelen
= getline(&line
, &len
, f
)) != -1) {
322 append_line(&contents
, &fulllen
, line
, linelen
);
327 drop_trailing_newlines(contents
);
332 static bool write_string(const char *fnam
, const char *string
)
337 if (!(f
= fopen(fnam
, "w")))
339 len
= strlen(string
);
340 ret
= fwrite(string
, 1, len
, f
);
342 fprintf(stderr
, "Error writing to file: %s\n", strerror(errno
));
347 fprintf(stderr
, "Error writing to file: %s\n", strerror(errno
));
360 static bool store_hierarchy(char *stridx
, char *h
)
362 if (num_hierarchies
% ALLOC_NUM
== 0) {
363 size_t n
= (num_hierarchies
/ ALLOC_NUM
) + 1;
365 char **tmp
= realloc(hierarchies
, n
* sizeof(char *));
367 fprintf(stderr
, "Out of memory\n");
373 hierarchies
[num_hierarchies
++] = must_copy_string(h
);
377 static void print_subsystems(void)
381 fprintf(stderr
, "hierarchies:\n");
382 for (i
= 0; i
< num_hierarchies
; i
++) {
384 fprintf(stderr
, " %d: %s\n", i
, hierarchies
[i
]);
388 static bool in_comma_list(const char *needle
, const char *haystack
)
390 const char *s
= haystack
, *e
;
391 size_t nlen
= strlen(needle
);
393 while (*s
&& (e
= index(s
, ','))) {
398 if (strncmp(needle
, s
, nlen
) == 0)
402 if (strcmp(needle
, s
) == 0)
407 /* do we need to do any massaging here? I'm not sure... */
408 /* Return the mounted controller and store the corresponding open file descriptor
409 * referring to the controller mountpoint in the private lxcfs namespace in
412 static char *find_mounted_controller(const char *controller
, int *cfd
)
416 for (i
= 0; i
< num_hierarchies
; i
++) {
419 if (strcmp(hierarchies
[i
], controller
) == 0) {
420 *cfd
= fd_hierarchies
[i
];
421 return hierarchies
[i
];
423 if (in_comma_list(controller
, hierarchies
[i
])) {
424 *cfd
= fd_hierarchies
[i
];
425 return hierarchies
[i
];
432 bool cgfs_set_value(const char *controller
, const char *cgroup
, const char *file
,
437 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
441 /* BASEDIR / tmpc / cgroup / file \0 */
442 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cgroup
) + strlen(file
) + 4;
444 snprintf(fnam
, len
, "%s/%s/%s/%s", BASEDIR
, tmpc
, cgroup
, file
);
446 return write_string(fnam
, value
);
449 // Chown all the files in the cgroup directory. We do this when we create
450 // a cgroup on behalf of a user.
451 static void chown_all_cgroup_files(const char *dirname
, uid_t uid
, gid_t gid
)
453 struct dirent dirent
, *direntp
;
454 char path
[MAXPATHLEN
];
459 len
= strlen(dirname
);
460 if (len
>= MAXPATHLEN
) {
461 fprintf(stderr
, "chown_all_cgroup_files: pathname too long: %s\n", dirname
);
465 d
= opendir(dirname
);
467 fprintf(stderr
, "chown_all_cgroup_files: failed to open %s\n", dirname
);
471 while (readdir_r(d
, &dirent
, &direntp
) == 0 && direntp
) {
472 if (!strcmp(direntp
->d_name
, ".") || !strcmp(direntp
->d_name
, ".."))
474 ret
= snprintf(path
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
475 if (ret
< 0 || ret
>= MAXPATHLEN
) {
476 fprintf(stderr
, "chown_all_cgroup_files: pathname too long under %s\n", dirname
);
479 if (chown(path
, uid
, gid
) < 0)
480 fprintf(stderr
, "Failed to chown file %s to %u:%u", path
, uid
, gid
);
485 int cgfs_create(const char *controller
, const char *cg
, uid_t uid
, gid_t gid
)
489 char *dirnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
493 /* BASEDIR / tmpc / cg \0 */
494 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cg
) + 3;
495 dirnam
= alloca(len
);
496 snprintf(dirnam
, len
, "%s/%s/%s", BASEDIR
,tmpc
, cg
);
498 if (mkdir(dirnam
, 0755) < 0)
501 if (uid
== 0 && gid
== 0)
504 if (chown(dirnam
, uid
, gid
) < 0)
507 chown_all_cgroup_files(dirnam
, uid
, gid
);
512 static bool recursive_rmdir(const char *dirname
)
514 struct dirent dirent
, *direntp
;
517 char pathname
[MAXPATHLEN
];
519 dir
= opendir(dirname
);
522 fprintf(stderr
, "%s: failed to open %s: %s\n", __func__
, dirname
, strerror(errno
));
527 while (!readdir_r(dir
, &dirent
, &direntp
)) {
534 if (!strcmp(direntp
->d_name
, ".") ||
535 !strcmp(direntp
->d_name
, ".."))
538 rc
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", dirname
, direntp
->d_name
);
539 if (rc
< 0 || rc
>= MAXPATHLEN
) {
540 fprintf(stderr
, "pathname too long\n");
544 ret
= lstat(pathname
, &mystat
);
547 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
551 if (S_ISDIR(mystat
.st_mode
)) {
552 if (!recursive_rmdir(pathname
)) {
554 fprintf(stderr
, "Error removing %s\n", pathname
);
561 if (closedir(dir
) < 0) {
562 fprintf(stderr
, "%s: failed to close directory %s: %s\n", __func__
, dirname
, strerror(errno
));
566 if (rmdir(dirname
) < 0) {
568 fprintf(stderr
, "%s: failed to delete %s: %s\n", __func__
, dirname
, strerror(errno
));
576 bool cgfs_remove(const char *controller
, const char *cg
)
580 char *dirnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
584 /* BASEDIR / tmpc / cg \0 */
585 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cg
) + 3;
586 dirnam
= alloca(len
);
587 snprintf(dirnam
, len
, "%s/%s/%s", BASEDIR
,tmpc
, cg
);
588 return recursive_rmdir(dirnam
);
591 bool cgfs_chmod_file(const char *controller
, const char *file
, mode_t mode
)
595 char *pathname
, *tmpc
= find_mounted_controller(controller
, &cfd
);
599 /* BASEDIR / tmpc / file \0 */
600 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(file
) + 3;
601 pathname
= alloca(len
);
602 snprintf(pathname
, len
, "%s/%s/%s", BASEDIR
, tmpc
, file
);
603 if (chmod(pathname
, mode
) < 0)
608 static int chown_tasks_files(const char *dirname
, uid_t uid
, gid_t gid
)
613 len
= strlen(dirname
) + strlen("/cgroup.procs") + 1;
615 snprintf(fname
, len
, "%s/tasks", dirname
);
616 if (chown(fname
, uid
, gid
) != 0)
618 snprintf(fname
, len
, "%s/cgroup.procs", dirname
);
619 if (chown(fname
, uid
, gid
) != 0)
624 int cgfs_chown_file(const char *controller
, const char *file
, uid_t uid
, gid_t gid
)
628 char *pathname
, *tmpc
= find_mounted_controller(controller
, &cfd
);
632 /* BASEDIR / tmpc / file \0 */
633 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(file
) + 3;
634 pathname
= alloca(len
);
635 snprintf(pathname
, len
, "%s/%s/%s", BASEDIR
, tmpc
, file
);
636 if (chown(pathname
, uid
, gid
) < 0)
639 if (is_dir(pathname
))
640 // like cgmanager did, we want to chown the tasks file as well
641 return chown_tasks_files(pathname
, uid
, gid
);
646 FILE *open_pids_file(const char *controller
, const char *cgroup
)
650 char *pathname
, *tmpc
= find_mounted_controller(controller
, &cfd
);
654 /* BASEDIR / tmpc / cgroup / "cgroup.procs" \0 */
655 len
= strlen(BASEDIR
) + strlen(tmpc
) + strlen(cgroup
) + 4 + strlen("cgroup.procs");
656 pathname
= alloca(len
);
657 snprintf(pathname
, len
, "%s/%s/%s/cgroup.procs", BASEDIR
, tmpc
, cgroup
);
658 return fopen(pathname
, "w");
661 static bool cgfs_iterate_cgroup(const char *controller
, const char *cgroup
, bool directories
,
662 void ***list
, size_t typesize
,
663 void* (*iterator
)(const char*, const char*, const char*))
668 char pathname
[MAXPATHLEN
];
669 size_t sz
= 0, asz
= 0;
670 struct dirent
*dirent
;
673 tmpc
= find_mounted_controller(controller
, &cfd
);
678 /* Make sure we pass a relative path to openat(). */
679 len
= strlen(cgroup
) + 1 /* . */ + 1 /* \0 */;
681 ret
= snprintf(cg
, len
, "%s%s", *cgroup
== '/' ? "." : "", cgroup
);
682 if (ret
< 0 || (size_t)ret
>= len
) {
683 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, cgroup
);
687 fd
= openat(cfd
, cg
, O_DIRECTORY
);
695 while ((dirent
= readdir(dir
))) {
698 if (!strcmp(dirent
->d_name
, ".") ||
699 !strcmp(dirent
->d_name
, ".."))
702 ret
= snprintf(pathname
, MAXPATHLEN
, "%s/%s", cg
, dirent
->d_name
);
703 if (ret
< 0 || ret
>= MAXPATHLEN
) {
704 fprintf(stderr
, "%s: pathname too long under %s\n", __func__
, cg
);
708 ret
= fstatat(cfd
, pathname
, &mystat
, AT_SYMLINK_NOFOLLOW
);
710 fprintf(stderr
, "%s: failed to stat %s: %s\n", __func__
, pathname
, strerror(errno
));
713 if ((!directories
&& !S_ISREG(mystat
.st_mode
)) ||
714 (directories
&& !S_ISDIR(mystat
.st_mode
)))
721 tmp
= realloc(*list
, asz
* typesize
);
725 (*list
)[sz
] = (*iterator
)(controller
, cg
, dirent
->d_name
);
726 (*list
)[sz
+1] = NULL
;
729 if (closedir(dir
) < 0) {
730 fprintf(stderr
, "%s: failed closedir for %s: %s\n", __func__
, cgroup
, strerror(errno
));
736 static void *make_children_list_entry(const char *controller
, const char *cgroup
, const char *dir_entry
)
740 dup
= strdup(dir_entry
);
745 bool cgfs_list_children(const char *controller
, const char *cgroup
, char ***list
)
747 return cgfs_iterate_cgroup(controller
, cgroup
, true, (void***)list
, sizeof(*list
), &make_children_list_entry
);
750 void free_key(struct cgfs_files
*k
)
758 void free_keys(struct cgfs_files
**keys
)
764 for (i
= 0; keys
[i
]; i
++) {
770 bool cgfs_get_value(const char *controller
, const char *cgroup
, const char *file
, char **value
)
774 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
778 /* . + /cgroup + / + file + \0 */
779 len
= strlen(cgroup
) + strlen(file
) + 3;
781 ret
= snprintf(fnam
, len
, "%s%s/%s", *cgroup
== '/' ? "." : "", cgroup
, file
);
782 if (ret
< 0 || (size_t)ret
>= len
)
785 fd
= openat(cfd
, fnam
, O_RDONLY
);
789 *value
= slurp_file(fnam
, fd
);
790 return *value
!= NULL
;
793 struct cgfs_files
*cgfs_get_key(const char *controller
, const char *cgroup
, const char *file
)
797 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
799 struct cgfs_files
*newkey
;
804 if (file
&& *file
== '/')
807 if (file
&& index(file
, '/'))
810 /* . + /cgroup + / + file + \0 */
811 len
= strlen(cgroup
) + 3;
813 len
+= strlen(file
) + 1;
815 snprintf(fnam
, len
, "%s%s%s%s", *cgroup
== '/' ? "." : "", cgroup
,
816 file
? "/" : "", file
? file
: "");
818 ret
= fstatat(cfd
, fnam
, &sb
, 0);
823 newkey
= malloc(sizeof(struct cgfs_files
));
826 newkey
->name
= must_copy_string(file
);
827 else if (rindex(cgroup
, '/'))
828 newkey
->name
= must_copy_string(rindex(cgroup
, '/'));
830 newkey
->name
= must_copy_string(cgroup
);
831 newkey
->uid
= sb
.st_uid
;
832 newkey
->gid
= sb
.st_gid
;
833 newkey
->mode
= sb
.st_mode
;
838 static void *make_key_list_entry(const char *controller
, const char *cgroup
, const char *dir_entry
)
840 struct cgfs_files
*entry
= cgfs_get_key(controller
, cgroup
, dir_entry
);
842 fprintf(stderr
, "%s: Error getting files under %s:%s\n",
843 __func__
, controller
, cgroup
);
848 bool cgfs_list_keys(const char *controller
, const char *cgroup
, struct cgfs_files
***keys
)
850 return cgfs_iterate_cgroup(controller
, cgroup
, false, (void***)keys
, sizeof(*keys
), &make_key_list_entry
);
853 bool is_child_cgroup(const char *controller
, const char *cgroup
, const char *f
)
857 char *fnam
, *tmpc
= find_mounted_controller(controller
, &cfd
);
863 /* . + /cgroup + / + f + \0 */
864 len
= strlen(cgroup
) + strlen(f
) + 3;
866 ret
= snprintf(fnam
, len
, "%s%s/%s", *cgroup
== '/' ? "." : "", cgroup
, f
);
867 if (ret
< 0 || (size_t)ret
>= len
)
870 ret
= fstatat(cfd
, fnam
, &sb
, 0);
871 if (ret
< 0 || !S_ISDIR(sb
.st_mode
))
876 #define SEND_CREDS_OK 0
877 #define SEND_CREDS_NOTSK 1
878 #define SEND_CREDS_FAIL 2
879 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
);
880 static int wait_for_pid(pid_t pid
);
881 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
);
882 static int send_creds_clone_wrapper(void *arg
);
885 * clone a task which switches to @task's namespace and writes '1'.
886 * over a unix sock so we can read the task's reaper's pid in our
889 * Note: glibc's fork() does not respect pidns, which can lead to failed
890 * assertions inside glibc (and thus failed forks) if the child's pid in
891 * the pidns and the parent pid outside are identical. Using clone prevents
894 static void write_task_init_pid_exit(int sock
, pid_t target
)
899 size_t stack_size
= sysconf(_SC_PAGESIZE
);
900 void *stack
= alloca(stack_size
);
902 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", (int)target
);
903 if (ret
< 0 || ret
>= sizeof(fnam
))
906 fd
= open(fnam
, O_RDONLY
);
908 perror("write_task_init_pid_exit open of ns/pid");
912 perror("write_task_init_pid_exit setns 1");
916 pid
= clone(send_creds_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &sock
);
920 if (!wait_for_pid(pid
))
926 static int send_creds_clone_wrapper(void *arg
) {
929 int sock
= *(int *)arg
;
931 /* we are the child */
936 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
)
941 static pid_t
get_init_pid_for_task(pid_t task
)
949 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
950 perror("socketpair");
959 write_task_init_pid_exit(sock
[0], task
);
963 if (!recv_creds(sock
[1], &cred
, &v
))
975 static pid_t
lookup_initpid_in_store(pid_t qpid
)
979 struct pidns_init_store
*e
;
982 snprintf(fnam
, 100, "/proc/%d/ns/pid", qpid
);
984 if (stat(fnam
, &sb
) < 0)
986 e
= lookup_verify_initpid(&sb
);
991 answer
= get_init_pid_for_task(qpid
);
993 save_initpid(&sb
, answer
);
996 /* we prune at end in case we are returning
997 * the value we were about to return */
998 prune_initpid_store();
1003 static int wait_for_pid(pid_t pid
)
1011 ret
= waitpid(pid
, &status
, 0);
1019 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
1026 * append pid to *src.
1027 * src: a pointer to a char* in which ot append the pid.
1028 * sz: the number of characters printed so far, minus trailing \0.
1029 * asz: the allocated size so far
1030 * pid: the pid to append
1032 static void must_strcat_pid(char **src
, size_t *sz
, size_t *asz
, pid_t pid
)
1036 int tmplen
= sprintf(tmp
, "%d\n", (int)pid
);
1038 if (!*src
|| tmplen
+ *sz
+ 1 >= *asz
) {
1041 tmp
= realloc(*src
, *asz
+ BUF_RESERVE_SIZE
);
1044 *asz
+= BUF_RESERVE_SIZE
;
1046 memcpy((*src
) +*sz
, tmp
, tmplen
+1); /* include the \0 */
1051 * Given a open file * to /proc/pid/{u,g}id_map, and an id
1052 * valid in the caller's namespace, return the id mapped into
1054 * Returns the mapped id, or -1 on error.
1057 convert_id_to_ns(FILE *idfile
, unsigned int in_id
)
1059 unsigned int nsuid
, // base id for a range in the idfile's namespace
1060 hostuid
, // base id for a range in the caller's namespace
1061 count
; // number of ids in this range
1065 fseek(idfile
, 0L, SEEK_SET
);
1066 while (fgets(line
, 400, idfile
)) {
1067 ret
= sscanf(line
, "%u %u %u\n", &nsuid
, &hostuid
, &count
);
1070 if (hostuid
+ count
< hostuid
|| nsuid
+ count
< nsuid
) {
1072 * uids wrapped around - unexpected as this is a procfile,
1075 fprintf(stderr
, "pid wrapparound at entry %u %u %u in %s\n",
1076 nsuid
, hostuid
, count
, line
);
1079 if (hostuid
<= in_id
&& hostuid
+count
> in_id
) {
1081 * now since hostuid <= in_id < hostuid+count, and
1082 * hostuid+count and nsuid+count do not wrap around,
1083 * we know that nsuid+(in_id-hostuid) which must be
1084 * less that nsuid+(count) must not wrap around
1086 return (in_id
- hostuid
) + nsuid
;
1095 * for is_privileged_over,
1096 * specify whether we require the calling uid to be root in his
1099 #define NS_ROOT_REQD true
1100 #define NS_ROOT_OPT false
1104 static bool is_privileged_over(pid_t pid
, uid_t uid
, uid_t victim
, bool req_ns_root
)
1106 char fpath
[PROCLEN
];
1108 bool answer
= false;
1111 if (victim
== -1 || uid
== -1)
1115 * If the request is one not requiring root in the namespace,
1116 * then having the same uid suffices. (i.e. uid 1000 has write
1117 * access to files owned by uid 1000
1119 if (!req_ns_root
&& uid
== victim
)
1122 ret
= snprintf(fpath
, PROCLEN
, "/proc/%d/uid_map", pid
);
1123 if (ret
< 0 || ret
>= PROCLEN
)
1125 FILE *f
= fopen(fpath
, "r");
1129 /* if caller's not root in his namespace, reject */
1130 nsuid
= convert_id_to_ns(f
, uid
);
1135 * If victim is not mapped into caller's ns, reject.
1136 * XXX I'm not sure this check is needed given that fuse
1137 * will be sending requests where the vfs has converted
1139 nsuid
= convert_id_to_ns(f
, victim
);
1150 static bool perms_include(int fmode
, mode_t req_mode
)
1154 switch (req_mode
& O_ACCMODE
) {
1162 r
= S_IROTH
| S_IWOTH
;
1167 return ((fmode
& r
) == r
);
1173 * querycg is /a/b/c/d/e
1176 static char *get_next_cgroup_dir(const char *taskcg
, const char *querycg
)
1180 if (strlen(taskcg
) <= strlen(querycg
)) {
1181 fprintf(stderr
, "%s: I was fed bad input\n", __func__
);
1185 if (strcmp(querycg
, "/") == 0)
1186 start
= strdup(taskcg
+ 1);
1188 start
= strdup(taskcg
+ strlen(querycg
) + 1);
1191 end
= strchr(start
, '/');
1197 static void stripnewline(char *x
)
1199 size_t l
= strlen(x
);
1200 if (l
&& x
[l
-1] == '\n')
1204 static char *get_pid_cgroup(pid_t pid
, const char *contrl
)
1209 char *answer
= NULL
;
1213 const char *h
= find_mounted_controller(contrl
, &cfd
);
1217 ret
= snprintf(fnam
, PROCLEN
, "/proc/%d/cgroup", pid
);
1218 if (ret
< 0 || ret
>= PROCLEN
)
1220 if (!(f
= fopen(fnam
, "r")))
1223 while (getline(&line
, &len
, f
) != -1) {
1227 c1
= strchr(line
, ':');
1231 c2
= strchr(c1
, ':');
1235 if (strcmp(c1
, h
) != 0)
1240 answer
= strdup(c2
);
1252 * check whether a fuse context may access a cgroup dir or file
1254 * If file is not null, it is a cgroup file to check under cg.
1255 * If file is null, then we are checking perms on cg itself.
1257 * For files we can check the mode of the list_keys result.
1258 * For cgroups, we must make assumptions based on the files under the
1259 * cgroup, because cgmanager doesn't tell us ownership/perms of cgroups
1262 static bool fc_may_access(struct fuse_context
*fc
, const char *contrl
, const char *cg
, const char *file
, mode_t mode
)
1264 struct cgfs_files
*k
= NULL
;
1267 k
= cgfs_get_key(contrl
, cg
, file
);
1271 if (is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
1272 if (perms_include(k
->mode
>> 6, mode
)) {
1277 if (fc
->gid
== k
->gid
) {
1278 if (perms_include(k
->mode
>> 3, mode
)) {
1283 ret
= perms_include(k
->mode
, mode
);
1290 #define INITSCOPE "/init.scope"
1291 static void prune_init_slice(char *cg
)
1294 size_t cg_len
= strlen(cg
), initscope_len
= strlen(INITSCOPE
);
1296 if (cg_len
< initscope_len
)
1299 point
= cg
+ cg_len
- initscope_len
;
1300 if (strcmp(point
, INITSCOPE
) == 0) {
1309 * If pid is in /a/b/c/d, he may only act on things under cg=/a/b/c/d.
1310 * If pid is in /a, he may act on /a/b, but not on /b.
1311 * if the answer is false and nextcg is not NULL, then *nextcg will point
1312 * to a string containing the next cgroup directory under cg, which must be
1313 * freed by the caller.
1315 static bool caller_is_in_ancestor(pid_t pid
, const char *contrl
, const char *cg
, char **nextcg
)
1317 bool answer
= false;
1318 char *c2
= get_pid_cgroup(pid
, contrl
);
1323 prune_init_slice(c2
);
1326 * callers pass in '/' or './' (openat()) for root cgroup, otherwise
1327 * they pass in a cgroup without leading '/'
1329 * The original line here was:
1330 * linecmp = *cg == '/' ? c2 : c2+1;
1331 * TODO: I'm not sure why you'd want to increment when *cg != '/'?
1332 * Serge, do you know?
1334 if (*cg
== '/' || !strncmp(cg
, "./", 2))
1338 if (strncmp(linecmp
, cg
, strlen(linecmp
)) != 0) {
1340 *nextcg
= get_next_cgroup_dir(linecmp
, cg
);
1352 * If pid is in /a/b/c, he may see that /a exists, but not /b or /a/c.
1354 static bool caller_may_see_dir(pid_t pid
, const char *contrl
, const char *cg
)
1356 bool answer
= false;
1358 size_t target_len
, task_len
;
1360 if (strcmp(cg
, "/") == 0 || strcmp(cg
, "./") == 0)
1363 c2
= get_pid_cgroup(pid
, contrl
);
1366 prune_init_slice(c2
);
1369 target_len
= strlen(cg
);
1370 task_len
= strlen(task_cg
);
1371 if (task_len
== 0) {
1372 /* Task is in the root cg, it can see everything. This case is
1373 * not handled by the strmcps below, since they test for the
1374 * last /, but that is the first / that we've chopped off
1380 if (strcmp(cg
, task_cg
) == 0) {
1384 if (target_len
< task_len
) {
1385 /* looking up a parent dir */
1386 if (strncmp(task_cg
, cg
, target_len
) == 0 && task_cg
[target_len
] == '/')
1390 if (target_len
> task_len
) {
1391 /* looking up a child dir */
1392 if (strncmp(task_cg
, cg
, task_len
) == 0 && cg
[task_len
] == '/')
1403 * given /cgroup/freezer/a/b, return "freezer".
1404 * the returned char* should NOT be freed.
1406 static char *pick_controller_from_path(struct fuse_context
*fc
, const char *path
)
1409 char *contr
, *slash
;
1411 if (strlen(path
) < 9)
1413 if (*(path
+7) != '/')
1416 contr
= strdupa(p1
);
1419 slash
= strstr(contr
, "/");
1424 for (i
= 0; i
< num_hierarchies
; i
++) {
1425 if (hierarchies
[i
] && strcmp(hierarchies
[i
], contr
) == 0)
1426 return hierarchies
[i
];
1432 * Find the start of cgroup in /cgroup/controller/the/cgroup/path
1433 * Note that the returned value may include files (keynames) etc
1435 static const char *find_cgroup_in_path(const char *path
)
1439 if (strlen(path
) < 9)
1441 p1
= strstr(path
+8, "/");
1448 * split the last path element from the path in @cg.
1449 * @dir is newly allocated and should be freed, @last not
1451 static void get_cgdir_and_path(const char *cg
, char **dir
, char **last
)
1458 *last
= strrchr(cg
, '/');
1463 p
= strrchr(*dir
, '/');
1468 * FUSE ops for /cgroup
1471 int cg_getattr(const char *path
, struct stat
*sb
)
1473 struct timespec now
;
1474 struct fuse_context
*fc
= fuse_get_context();
1475 char * cgdir
= NULL
;
1476 char *last
= NULL
, *path1
, *path2
;
1477 struct cgfs_files
*k
= NULL
;
1479 const char *controller
= NULL
;
1486 memset(sb
, 0, sizeof(struct stat
));
1488 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
1491 sb
->st_uid
= sb
->st_gid
= 0;
1492 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
1495 if (strcmp(path
, "/cgroup") == 0) {
1496 sb
->st_mode
= S_IFDIR
| 00755;
1501 controller
= pick_controller_from_path(fc
, path
);
1504 cgroup
= find_cgroup_in_path(path
);
1506 /* this is just /cgroup/controller, return it as a dir */
1507 sb
->st_mode
= S_IFDIR
| 00755;
1512 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1522 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1525 /* check that cgcopy is either a child cgroup of cgdir, or listed in its keys.
1526 * Then check that caller's cgroup is under path if last is a child
1527 * cgroup, or cgdir if last is a file */
1529 if (is_child_cgroup(controller
, path1
, path2
)) {
1530 if (!caller_may_see_dir(initpid
, controller
, cgroup
)) {
1534 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
1535 /* this is just /cgroup/controller, return it as a dir */
1536 sb
->st_mode
= S_IFDIR
| 00555;
1541 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
)) {
1546 // get uid, gid, from '/tasks' file and make up a mode
1547 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
1548 sb
->st_mode
= S_IFDIR
| 00755;
1549 k
= cgfs_get_key(controller
, cgroup
, NULL
);
1551 sb
->st_uid
= sb
->st_gid
= 0;
1553 sb
->st_uid
= k
->uid
;
1554 sb
->st_gid
= k
->gid
;
1562 if ((k
= cgfs_get_key(controller
, path1
, path2
)) != NULL
) {
1563 sb
->st_mode
= S_IFREG
| k
->mode
;
1565 sb
->st_uid
= k
->uid
;
1566 sb
->st_gid
= k
->gid
;
1569 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
1573 if (!fc_may_access(fc
, controller
, path1
, path2
, O_RDONLY
)) {
1586 int cg_opendir(const char *path
, struct fuse_file_info
*fi
)
1588 struct fuse_context
*fc
= fuse_get_context();
1590 struct file_info
*dir_info
;
1591 char *controller
= NULL
;
1596 if (strcmp(path
, "/cgroup") == 0) {
1600 // return list of keys for the controller, and list of child cgroups
1601 controller
= pick_controller_from_path(fc
, path
);
1605 cgroup
= find_cgroup_in_path(path
);
1607 /* this is just /cgroup/controller, return its contents */
1612 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1616 if (!caller_may_see_dir(initpid
, controller
, cgroup
))
1618 if (!fc_may_access(fc
, controller
, cgroup
, NULL
, O_RDONLY
))
1622 /* we'll free this at cg_releasedir */
1623 dir_info
= malloc(sizeof(*dir_info
));
1626 dir_info
->controller
= must_copy_string(controller
);
1627 dir_info
->cgroup
= must_copy_string(cgroup
);
1628 dir_info
->type
= LXC_TYPE_CGDIR
;
1629 dir_info
->buf
= NULL
;
1630 dir_info
->file
= NULL
;
1631 dir_info
->buflen
= 0;
1633 fi
->fh
= (unsigned long)dir_info
;
1637 int cg_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
1638 struct fuse_file_info
*fi
)
1640 struct file_info
*d
= (struct file_info
*)fi
->fh
;
1641 struct cgfs_files
**list
= NULL
;
1643 char *nextcg
= NULL
;
1644 struct fuse_context
*fc
= fuse_get_context();
1645 char **clist
= NULL
;
1647 if (d
->type
!= LXC_TYPE_CGDIR
) {
1648 fprintf(stderr
, "Internal error: file cache info used in readdir\n");
1651 if (!d
->cgroup
&& !d
->controller
) {
1652 // ls /var/lib/lxcfs/cgroup - just show list of controllers
1655 for (i
= 0; i
< num_hierarchies
; i
++) {
1656 if (hierarchies
[i
] && filler(buf
, hierarchies
[i
], NULL
, 0) != 0) {
1663 if (!cgfs_list_keys(d
->controller
, d
->cgroup
, &list
)) {
1664 // not a valid cgroup
1669 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1672 if (!caller_is_in_ancestor(initpid
, d
->controller
, d
->cgroup
, &nextcg
)) {
1674 ret
= filler(buf
, nextcg
, NULL
, 0);
1685 for (i
= 0; list
[i
]; i
++) {
1686 if (filler(buf
, list
[i
]->name
, NULL
, 0) != 0) {
1692 // now get the list of child cgroups
1694 if (!cgfs_list_children(d
->controller
, d
->cgroup
, &clist
)) {
1699 for (i
= 0; clist
[i
]; i
++) {
1700 if (filler(buf
, clist
[i
], NULL
, 0) != 0) {
1711 for (i
= 0; clist
[i
]; i
++)
1718 static void do_release_file_info(struct fuse_file_info
*fi
)
1720 struct file_info
*f
= (struct file_info
*)fi
->fh
;
1727 free(f
->controller
);
1728 f
->controller
= NULL
;
1738 int cg_releasedir(const char *path
, struct fuse_file_info
*fi
)
1740 do_release_file_info(fi
);
1744 int cg_open(const char *path
, struct fuse_file_info
*fi
)
1747 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1748 struct cgfs_files
*k
= NULL
;
1749 struct file_info
*file_info
;
1750 struct fuse_context
*fc
= fuse_get_context();
1756 controller
= pick_controller_from_path(fc
, path
);
1759 cgroup
= find_cgroup_in_path(path
);
1763 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1772 k
= cgfs_get_key(controller
, path1
, path2
);
1779 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1782 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1786 if (!fc_may_access(fc
, controller
, path1
, path2
, fi
->flags
)) {
1791 /* we'll free this at cg_release */
1792 file_info
= malloc(sizeof(*file_info
));
1797 file_info
->controller
= must_copy_string(controller
);
1798 file_info
->cgroup
= must_copy_string(path1
);
1799 file_info
->file
= must_copy_string(path2
);
1800 file_info
->type
= LXC_TYPE_CGFILE
;
1801 file_info
->buf
= NULL
;
1802 file_info
->buflen
= 0;
1804 fi
->fh
= (unsigned long)file_info
;
1812 int cg_access(const char *path
, int mode
)
1815 char *last
= NULL
, *path1
, *path2
, * cgdir
= NULL
, *controller
;
1816 struct cgfs_files
*k
= NULL
;
1817 struct fuse_context
*fc
= fuse_get_context();
1823 controller
= pick_controller_from_path(fc
, path
);
1826 cgroup
= find_cgroup_in_path(path
);
1828 // access("/sys/fs/cgroup/systemd", mode) - rx allowed, w not
1829 if ((mode
& W_OK
) == 0)
1834 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
1843 k
= cgfs_get_key(controller
, path1
, path2
);
1845 if ((mode
& W_OK
) == 0)
1853 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
1856 if (!caller_may_see_dir(initpid
, controller
, path1
)) {
1860 if (!fc_may_access(fc
, controller
, path1
, path2
, mode
)) {
1872 int cg_release(const char *path
, struct fuse_file_info
*fi
)
1874 do_release_file_info(fi
);
1878 #define POLLIN_SET ( EPOLLIN | EPOLLHUP | EPOLLRDHUP )
1880 static bool wait_for_sock(int sock
, int timeout
)
1882 struct epoll_event ev
;
1883 int epfd
, ret
, now
, starttime
, deltatime
, saved_errno
;
1885 if ((starttime
= time(NULL
)) < 0)
1888 if ((epfd
= epoll_create(1)) < 0) {
1889 fprintf(stderr
, "Failed to create epoll socket: %m\n");
1893 ev
.events
= POLLIN_SET
;
1895 if (epoll_ctl(epfd
, EPOLL_CTL_ADD
, sock
, &ev
) < 0) {
1896 fprintf(stderr
, "Failed adding socket to epoll: %m\n");
1902 if ((now
= time(NULL
)) < 0) {
1907 deltatime
= (starttime
+ timeout
) - now
;
1908 if (deltatime
< 0) { // timeout
1913 ret
= epoll_wait(epfd
, &ev
, 1, 1000*deltatime
+ 1);
1914 if (ret
< 0 && errno
== EINTR
)
1916 saved_errno
= errno
;
1920 errno
= saved_errno
;
1926 static int msgrecv(int sockfd
, void *buf
, size_t len
)
1928 if (!wait_for_sock(sockfd
, 2))
1930 return recv(sockfd
, buf
, len
, MSG_DONTWAIT
);
1933 static int send_creds(int sock
, struct ucred
*cred
, char v
, bool pingfirst
)
1935 struct msghdr msg
= { 0 };
1937 struct cmsghdr
*cmsg
;
1938 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1943 if (msgrecv(sock
, buf
, 1) != 1) {
1944 fprintf(stderr
, "%s: Error getting reply from server over socketpair\n",
1946 return SEND_CREDS_FAIL
;
1950 msg
.msg_control
= cmsgbuf
;
1951 msg
.msg_controllen
= sizeof(cmsgbuf
);
1953 cmsg
= CMSG_FIRSTHDR(&msg
);
1954 cmsg
->cmsg_len
= CMSG_LEN(sizeof(struct ucred
));
1955 cmsg
->cmsg_level
= SOL_SOCKET
;
1956 cmsg
->cmsg_type
= SCM_CREDENTIALS
;
1957 memcpy(CMSG_DATA(cmsg
), cred
, sizeof(*cred
));
1959 msg
.msg_name
= NULL
;
1960 msg
.msg_namelen
= 0;
1964 iov
.iov_len
= sizeof(buf
);
1968 if (sendmsg(sock
, &msg
, 0) < 0) {
1969 fprintf(stderr
, "%s: failed at sendmsg: %s\n", __func__
,
1972 return SEND_CREDS_NOTSK
;
1973 return SEND_CREDS_FAIL
;
1976 return SEND_CREDS_OK
;
1979 static bool recv_creds(int sock
, struct ucred
*cred
, char *v
)
1981 struct msghdr msg
= { 0 };
1983 struct cmsghdr
*cmsg
;
1984 char cmsgbuf
[CMSG_SPACE(sizeof(*cred
))];
1995 if (setsockopt(sock
, SOL_SOCKET
, SO_PASSCRED
, &optval
, sizeof(optval
)) == -1) {
1996 fprintf(stderr
, "Failed to set passcred: %s\n", strerror(errno
));
2000 if (write(sock
, buf
, 1) != 1) {
2001 fprintf(stderr
, "Failed to start write on scm fd: %s\n", strerror(errno
));
2005 msg
.msg_name
= NULL
;
2006 msg
.msg_namelen
= 0;
2007 msg
.msg_control
= cmsgbuf
;
2008 msg
.msg_controllen
= sizeof(cmsgbuf
);
2011 iov
.iov_len
= sizeof(buf
);
2015 if (!wait_for_sock(sock
, 2)) {
2016 fprintf(stderr
, "Timed out waiting for scm_cred: %s\n",
2020 ret
= recvmsg(sock
, &msg
, MSG_DONTWAIT
);
2022 fprintf(stderr
, "Failed to receive scm_cred: %s\n",
2027 cmsg
= CMSG_FIRSTHDR(&msg
);
2029 if (cmsg
&& cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
)) &&
2030 cmsg
->cmsg_level
== SOL_SOCKET
&&
2031 cmsg
->cmsg_type
== SCM_CREDENTIALS
) {
2032 memcpy(cred
, CMSG_DATA(cmsg
), sizeof(*cred
));
2039 struct pid_ns_clone_args
{
2043 int (*wrapped
) (int, pid_t
); // pid_from_ns or pid_to_ns
2047 * pid_ns_clone_wrapper - wraps pid_to_ns or pid_from_ns for usage
2048 * with clone(). This simply writes '1' as ACK back to the parent
2049 * before calling the actual wrapped function.
2051 static int pid_ns_clone_wrapper(void *arg
) {
2052 struct pid_ns_clone_args
* args
= (struct pid_ns_clone_args
*) arg
;
2055 close(args
->cpipe
[0]);
2056 if (write(args
->cpipe
[1], &b
, sizeof(char)) < 0) {
2057 fprintf(stderr
, "%s (child): error on write: %s\n",
2058 __func__
, strerror(errno
));
2060 close(args
->cpipe
[1]);
2061 return args
->wrapped(args
->sock
, args
->tpid
);
2065 * pid_to_ns - reads pids from a ucred over a socket, then writes the
2066 * int value back over the socket. This shifts the pid from the
2067 * sender's pidns into tpid's pidns.
2069 static int pid_to_ns(int sock
, pid_t tpid
)
2074 while (recv_creds(sock
, &cred
, &v
)) {
2077 if (write(sock
, &cred
.pid
, sizeof(pid_t
)) != sizeof(pid_t
))
2085 * pid_to_ns_wrapper: when you setns into a pidns, you yourself remain
2086 * in your old pidns. Only children which you clone will be in the target
2087 * pidns. So the pid_to_ns_wrapper does the setns, then clones a child to
2088 * actually convert pids.
2090 * Note: glibc's fork() does not respect pidns, which can lead to failed
2091 * assertions inside glibc (and thus failed forks) if the child's pid in
2092 * the pidns and the parent pid outside are identical. Using clone prevents
2095 static void pid_to_ns_wrapper(int sock
, pid_t tpid
)
2097 int newnsfd
= -1, ret
, cpipe
[2];
2102 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2103 if (ret
< 0 || ret
>= sizeof(fnam
))
2105 newnsfd
= open(fnam
, O_RDONLY
);
2108 if (setns(newnsfd
, 0) < 0)
2112 if (pipe(cpipe
) < 0)
2115 struct pid_ns_clone_args args
= {
2119 .wrapped
= &pid_to_ns
2121 size_t stack_size
= sysconf(_SC_PAGESIZE
);
2122 void *stack
= alloca(stack_size
);
2124 cpid
= clone(pid_ns_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &args
);
2128 // give the child 1 second to be done forking and
2130 if (!wait_for_sock(cpipe
[0], 1))
2132 ret
= read(cpipe
[0], &v
, 1);
2133 if (ret
!= sizeof(char) || v
!= '1')
2136 if (!wait_for_pid(cpid
))
2142 * To read cgroup files with a particular pid, we will setns into the child
2143 * pidns, open a pipe, fork a child - which will be the first to really be in
2144 * the child ns - which does the cgfs_get_value and writes the data to the pipe.
2146 bool do_read_pids(pid_t tpid
, const char *contrl
, const char *cg
, const char *file
, char **d
)
2148 int sock
[2] = {-1, -1};
2149 char *tmpdata
= NULL
;
2151 pid_t qpid
, cpid
= -1;
2152 bool answer
= false;
2155 size_t sz
= 0, asz
= 0;
2157 if (!cgfs_get_value(contrl
, cg
, file
, &tmpdata
))
2161 * Now we read the pids from returned data one by one, pass
2162 * them into a child in the target namespace, read back the
2163 * translated pids, and put them into our to-return data
2166 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2167 perror("socketpair");
2176 if (!cpid
) // child - exits when done
2177 pid_to_ns_wrapper(sock
[1], tpid
);
2179 char *ptr
= tmpdata
;
2182 while (sscanf(ptr
, "%d\n", &qpid
) == 1) {
2184 ret
= send_creds(sock
[0], &cred
, v
, true);
2186 if (ret
== SEND_CREDS_NOTSK
)
2188 if (ret
== SEND_CREDS_FAIL
)
2191 // read converted results
2192 if (!wait_for_sock(sock
[0], 2)) {
2193 fprintf(stderr
, "%s: timed out waiting for pid from child: %s\n",
2194 __func__
, strerror(errno
));
2197 if (read(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2198 fprintf(stderr
, "%s: error reading pid from child: %s\n",
2199 __func__
, strerror(errno
));
2202 must_strcat_pid(d
, &sz
, &asz
, qpid
);
2204 ptr
= strchr(ptr
, '\n');
2210 cred
.pid
= getpid();
2212 if (send_creds(sock
[0], &cred
, v
, true) != SEND_CREDS_OK
) {
2213 // failed to ask child to exit
2214 fprintf(stderr
, "%s: failed to ask child to exit: %s\n",
2215 __func__
, strerror(errno
));
2225 if (sock
[0] != -1) {
2232 int cg_read(const char *path
, char *buf
, size_t size
, off_t offset
,
2233 struct fuse_file_info
*fi
)
2235 struct fuse_context
*fc
= fuse_get_context();
2236 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2237 struct cgfs_files
*k
= NULL
;
2242 if (f
->type
!= LXC_TYPE_CGFILE
) {
2243 fprintf(stderr
, "Internal error: directory cache info used in cg_read\n");
2256 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2262 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_RDONLY
)) {
2267 if (strcmp(f
->file
, "tasks") == 0 ||
2268 strcmp(f
->file
, "/tasks") == 0 ||
2269 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2270 strcmp(f
->file
, "cgroup.procs") == 0)
2271 // special case - we have to translate the pids
2272 r
= do_read_pids(fc
->pid
, f
->controller
, f
->cgroup
, f
->file
, &data
);
2274 r
= cgfs_get_value(f
->controller
, f
->cgroup
, f
->file
, &data
);
2288 memcpy(buf
, data
, s
);
2289 if (s
> 0 && s
< size
&& data
[s
-1] != '\n')
2299 static int pid_from_ns(int sock
, pid_t tpid
)
2309 if (!wait_for_sock(sock
, 2)) {
2310 fprintf(stderr
, "%s: timeout reading from parent\n", __func__
);
2313 if ((ret
= read(sock
, &vpid
, sizeof(pid_t
))) != sizeof(pid_t
)) {
2314 fprintf(stderr
, "%s: bad read from parent: %s\n",
2315 __func__
, strerror(errno
));
2318 if (vpid
== -1) // done
2322 if (send_creds(sock
, &cred
, v
, true) != SEND_CREDS_OK
) {
2324 cred
.pid
= getpid();
2325 if (send_creds(sock
, &cred
, v
, false) != SEND_CREDS_OK
)
2332 static void pid_from_ns_wrapper(int sock
, pid_t tpid
)
2334 int newnsfd
= -1, ret
, cpipe
[2];
2339 ret
= snprintf(fnam
, sizeof(fnam
), "/proc/%d/ns/pid", tpid
);
2340 if (ret
< 0 || ret
>= sizeof(fnam
))
2342 newnsfd
= open(fnam
, O_RDONLY
);
2345 if (setns(newnsfd
, 0) < 0)
2349 if (pipe(cpipe
) < 0)
2352 struct pid_ns_clone_args args
= {
2356 .wrapped
= &pid_from_ns
2358 size_t stack_size
= sysconf(_SC_PAGESIZE
);
2359 void *stack
= alloca(stack_size
);
2361 cpid
= clone(pid_ns_clone_wrapper
, stack
+ stack_size
, SIGCHLD
, &args
);
2365 // give the child 1 second to be done forking and
2367 if (!wait_for_sock(cpipe
[0], 1))
2369 ret
= read(cpipe
[0], &v
, 1);
2370 if (ret
!= sizeof(char) || v
!= '1')
2373 if (!wait_for_pid(cpid
))
2379 * Given host @uid, return the uid to which it maps in
2380 * @pid's user namespace, or -1 if none.
2382 bool hostuid_to_ns(uid_t uid
, pid_t pid
, uid_t
*answer
)
2387 sprintf(line
, "/proc/%d/uid_map", pid
);
2388 if ((f
= fopen(line
, "r")) == NULL
) {
2392 *answer
= convert_id_to_ns(f
, uid
);
2401 * get_pid_creds: get the real uid and gid of @pid from
2403 * (XXX should we use euid here?)
2405 void get_pid_creds(pid_t pid
, uid_t
*uid
, gid_t
*gid
)
2414 sprintf(line
, "/proc/%d/status", pid
);
2415 if ((f
= fopen(line
, "r")) == NULL
) {
2416 fprintf(stderr
, "Error opening %s: %s\n", line
, strerror(errno
));
2419 while (fgets(line
, 400, f
)) {
2420 if (strncmp(line
, "Uid:", 4) == 0) {
2421 if (sscanf(line
+4, "%u", &u
) != 1) {
2422 fprintf(stderr
, "bad uid line for pid %u\n", pid
);
2427 } else if (strncmp(line
, "Gid:", 4) == 0) {
2428 if (sscanf(line
+4, "%u", &g
) != 1) {
2429 fprintf(stderr
, "bad gid line for pid %u\n", pid
);
2440 * May the requestor @r move victim @v to a new cgroup?
2441 * This is allowed if
2442 * . they are the same task
2443 * . they are ownedy by the same uid
2444 * . @r is root on the host, or
2445 * . @v's uid is mapped into @r's where @r is root.
2447 bool may_move_pid(pid_t r
, uid_t r_uid
, pid_t v
)
2449 uid_t v_uid
, tmpuid
;
2456 get_pid_creds(v
, &v_uid
, &v_gid
);
2459 if (hostuid_to_ns(r_uid
, r
, &tmpuid
) && tmpuid
== 0
2460 && hostuid_to_ns(v_uid
, r
, &tmpuid
))
2465 static bool do_write_pids(pid_t tpid
, uid_t tuid
, const char *contrl
, const char *cg
,
2466 const char *file
, const char *buf
)
2468 int sock
[2] = {-1, -1};
2469 pid_t qpid
, cpid
= -1;
2470 FILE *pids_file
= NULL
;
2471 bool answer
= false, fail
= false;
2473 pids_file
= open_pids_file(contrl
, cg
);
2478 * write the pids to a socket, have helper in writer's pidns
2479 * call movepid for us
2481 if (socketpair(AF_UNIX
, SOCK_DGRAM
, 0, sock
) < 0) {
2482 perror("socketpair");
2490 if (!cpid
) { // child
2492 pid_from_ns_wrapper(sock
[1], tpid
);
2495 const char *ptr
= buf
;
2496 while (sscanf(ptr
, "%d", &qpid
) == 1) {
2500 if (write(sock
[0], &qpid
, sizeof(qpid
)) != sizeof(qpid
)) {
2501 fprintf(stderr
, "%s: error writing pid to child: %s\n",
2502 __func__
, strerror(errno
));
2506 if (recv_creds(sock
[0], &cred
, &v
)) {
2508 if (!may_move_pid(tpid
, tuid
, cred
.pid
)) {
2512 if (fprintf(pids_file
, "%d", (int) cred
.pid
) < 0)
2517 ptr
= strchr(ptr
, '\n');
2523 /* All good, write the value */
2525 if (write(sock
[0], &qpid
,sizeof(qpid
)) != sizeof(qpid
))
2526 fprintf(stderr
, "Warning: failed to ask child to exit\n");
2534 if (sock
[0] != -1) {
2539 if (fclose(pids_file
) != 0)
2545 int cg_write(const char *path
, const char *buf
, size_t size
, off_t offset
,
2546 struct fuse_file_info
*fi
)
2548 struct fuse_context
*fc
= fuse_get_context();
2549 char *localbuf
= NULL
;
2550 struct cgfs_files
*k
= NULL
;
2551 struct file_info
*f
= (struct file_info
*)fi
->fh
;
2554 if (f
->type
!= LXC_TYPE_CGFILE
) {
2555 fprintf(stderr
, "Internal error: directory cache info used in cg_write\n");
2565 localbuf
= alloca(size
+1);
2566 localbuf
[size
] = '\0';
2567 memcpy(localbuf
, buf
, size
);
2569 if ((k
= cgfs_get_key(f
->controller
, f
->cgroup
, f
->file
)) == NULL
) {
2574 if (!fc_may_access(fc
, f
->controller
, f
->cgroup
, f
->file
, O_WRONLY
)) {
2579 if (strcmp(f
->file
, "tasks") == 0 ||
2580 strcmp(f
->file
, "/tasks") == 0 ||
2581 strcmp(f
->file
, "/cgroup.procs") == 0 ||
2582 strcmp(f
->file
, "cgroup.procs") == 0)
2583 // special case - we have to translate the pids
2584 r
= do_write_pids(fc
->pid
, fc
->uid
, f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2586 r
= cgfs_set_value(f
->controller
, f
->cgroup
, f
->file
, localbuf
);
2596 int cg_chown(const char *path
, uid_t uid
, gid_t gid
)
2598 struct fuse_context
*fc
= fuse_get_context();
2599 char *cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2600 struct cgfs_files
*k
= NULL
;
2607 if (strcmp(path
, "/cgroup") == 0)
2610 controller
= pick_controller_from_path(fc
, path
);
2613 cgroup
= find_cgroup_in_path(path
);
2615 /* this is just /cgroup/controller */
2618 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2628 if (is_child_cgroup(controller
, path1
, path2
)) {
2629 // get uid, gid, from '/tasks' file and make up a mode
2630 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2631 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2634 k
= cgfs_get_key(controller
, path1
, path2
);
2642 * This being a fuse request, the uid and gid must be valid
2643 * in the caller's namespace. So we can just check to make
2644 * sure that the caller is root in his uid, and privileged
2645 * over the file's current owner.
2647 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_REQD
)) {
2652 ret
= cgfs_chown_file(controller
, cgroup
, uid
, gid
);
2661 int cg_chmod(const char *path
, mode_t mode
)
2663 struct fuse_context
*fc
= fuse_get_context();
2664 char * cgdir
= NULL
, *last
= NULL
, *path1
, *path2
, *controller
;
2665 struct cgfs_files
*k
= NULL
;
2672 if (strcmp(path
, "/cgroup") == 0)
2675 controller
= pick_controller_from_path(fc
, path
);
2678 cgroup
= find_cgroup_in_path(path
);
2680 /* this is just /cgroup/controller */
2683 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2693 if (is_child_cgroup(controller
, path1
, path2
)) {
2694 // get uid, gid, from '/tasks' file and make up a mode
2695 // That is a hack, until cgmanager gains a GetCgroupPerms fn.
2696 k
= cgfs_get_key(controller
, cgroup
, "tasks");
2699 k
= cgfs_get_key(controller
, path1
, path2
);
2707 * This being a fuse request, the uid and gid must be valid
2708 * in the caller's namespace. So we can just check to make
2709 * sure that the caller is root in his uid, and privileged
2710 * over the file's current owner.
2712 if (!is_privileged_over(fc
->pid
, fc
->uid
, k
->uid
, NS_ROOT_OPT
)) {
2717 if (!cgfs_chmod_file(controller
, cgroup
, mode
)) {
2729 int cg_mkdir(const char *path
, mode_t mode
)
2731 struct fuse_context
*fc
= fuse_get_context();
2732 char *last
= NULL
, *path1
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2740 controller
= pick_controller_from_path(fc
, path
);
2744 cgroup
= find_cgroup_in_path(path
);
2748 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2754 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2757 if (!caller_is_in_ancestor(initpid
, controller
, path1
, &next
)) {
2760 else if (last
&& strcmp(next
, last
) == 0)
2767 if (!fc_may_access(fc
, controller
, path1
, NULL
, O_RDWR
)) {
2771 if (!caller_is_in_ancestor(initpid
, controller
, path1
, NULL
)) {
2776 ret
= cgfs_create(controller
, cgroup
, fc
->uid
, fc
->gid
);
2784 int cg_rmdir(const char *path
)
2786 struct fuse_context
*fc
= fuse_get_context();
2787 char *last
= NULL
, *cgdir
= NULL
, *controller
, *next
= NULL
;
2794 controller
= pick_controller_from_path(fc
, path
);
2798 cgroup
= find_cgroup_in_path(path
);
2802 get_cgdir_and_path(cgroup
, &cgdir
, &last
);
2808 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2811 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, &next
)) {
2812 if (!last
|| strcmp(next
, last
) == 0)
2819 if (!fc_may_access(fc
, controller
, cgdir
, NULL
, O_WRONLY
)) {
2823 if (!caller_is_in_ancestor(initpid
, controller
, cgroup
, NULL
)) {
2828 if (!cgfs_remove(controller
, cgroup
)) {
2841 static bool startswith(const char *line
, const char *pref
)
2843 if (strncmp(line
, pref
, strlen(pref
)) == 0)
2848 static void get_mem_cached(char *memstat
, unsigned long *v
)
2854 if (startswith(memstat
, "total_cache")) {
2855 sscanf(memstat
+ 11, "%lu", v
);
2859 eol
= strchr(memstat
, '\n');
2866 static void get_blkio_io_value(char *str
, unsigned major
, unsigned minor
, char *iotype
, unsigned long *v
)
2872 snprintf(key
, 32, "%u:%u %s", major
, minor
, iotype
);
2874 size_t len
= strlen(key
);
2878 if (startswith(str
, key
)) {
2879 sscanf(str
+ len
, "%lu", v
);
2882 eol
= strchr(str
, '\n');
2889 static int read_file(const char *path
, char *buf
, size_t size
,
2890 struct file_info
*d
)
2892 size_t linelen
= 0, total_len
= 0, rv
= 0;
2894 char *cache
= d
->buf
;
2895 size_t cache_size
= d
->buflen
;
2896 FILE *f
= fopen(path
, "r");
2900 while (getline(&line
, &linelen
, f
) != -1) {
2901 ssize_t l
= snprintf(cache
, cache_size
, "%s", line
);
2903 perror("Error writing to cache");
2907 if (l
>= cache_size
) {
2908 fprintf(stderr
, "Internal error: truncated write to cache\n");
2917 d
->size
= total_len
;
2918 if (total_len
> size
)
2921 /* read from off 0 */
2922 memcpy(buf
, d
->buf
, total_len
);
2931 * FUSE ops for /proc
2934 static unsigned long get_memlimit(const char *cgroup
)
2936 char *memlimit_str
= NULL
;
2937 unsigned long memlimit
= -1;
2939 if (cgfs_get_value("memory", cgroup
, "memory.limit_in_bytes", &memlimit_str
))
2940 memlimit
= strtoul(memlimit_str
, NULL
, 10);
2947 static unsigned long get_min_memlimit(const char *cgroup
)
2949 char *copy
= strdupa(cgroup
);
2950 unsigned long memlimit
= 0, retlimit
;
2952 retlimit
= get_memlimit(copy
);
2954 while (strcmp(copy
, "/") != 0) {
2955 copy
= dirname(copy
);
2956 memlimit
= get_memlimit(copy
);
2957 if (memlimit
!= -1 && memlimit
< retlimit
)
2958 retlimit
= memlimit
;
2964 static int proc_meminfo_read(char *buf
, size_t size
, off_t offset
,
2965 struct fuse_file_info
*fi
)
2967 struct fuse_context
*fc
= fuse_get_context();
2968 struct file_info
*d
= (struct file_info
*)fi
->fh
;
2970 char *memusage_str
= NULL
, *memstat_str
= NULL
,
2971 *memswlimit_str
= NULL
, *memswusage_str
= NULL
,
2972 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
2973 unsigned long memlimit
= 0, memusage
= 0, memswlimit
= 0, memswusage
= 0,
2974 cached
= 0, hosttotal
= 0;
2976 size_t linelen
= 0, total_len
= 0, rv
= 0;
2977 char *cache
= d
->buf
;
2978 size_t cache_size
= d
->buflen
;
2982 if (offset
> d
->size
)
2986 int left
= d
->size
- offset
;
2987 total_len
= left
> size
? size
: left
;
2988 memcpy(buf
, cache
+ offset
, total_len
);
2992 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
2995 cg
= get_pid_cgroup(initpid
, "memory");
2997 return read_file("/proc/meminfo", buf
, size
, d
);
2998 prune_init_slice(cg
);
3000 memlimit
= get_min_memlimit(cg
);
3001 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
3003 if (!cgfs_get_value("memory", cg
, "memory.stat", &memstat_str
))
3006 // Following values are allowed to fail, because swapaccount might be turned
3007 // off for current kernel
3008 if(cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
) &&
3009 cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
))
3011 /* If swapaccounting is turned on, then default value is assumed to be that of cgroup / */
3012 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
3014 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
3017 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
3018 memswusage
= strtoul(memswusage_str
, NULL
, 10);
3020 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
3022 if (!strcmp(memswusage_str
, memswusage_default_str
))
3025 memswlimit
= memswlimit
/ 1024;
3026 memswusage
= memswusage
/ 1024;
3029 memusage
= strtoul(memusage_str
, NULL
, 10);
3033 get_mem_cached(memstat_str
, &cached
);
3035 f
= fopen("/proc/meminfo", "r");
3039 while (getline(&line
, &linelen
, f
) != -1) {
3041 char *printme
, lbuf
[100];
3043 memset(lbuf
, 0, 100);
3044 if (startswith(line
, "MemTotal:")) {
3045 sscanf(line
+14, "%lu", &hosttotal
);
3046 if (hosttotal
< memlimit
)
3047 memlimit
= hosttotal
;
3048 snprintf(lbuf
, 100, "MemTotal: %8lu kB\n", memlimit
);
3050 } else if (startswith(line
, "MemFree:")) {
3051 snprintf(lbuf
, 100, "MemFree: %8lu kB\n", memlimit
- memusage
);
3053 } else if (startswith(line
, "MemAvailable:")) {
3054 snprintf(lbuf
, 100, "MemAvailable: %8lu kB\n", memlimit
- memusage
);
3056 } else if (startswith(line
, "SwapTotal:") && memswlimit
> 0) {
3057 snprintf(lbuf
, 100, "SwapTotal: %8lu kB\n", memswlimit
- memlimit
);
3059 } else if (startswith(line
, "SwapFree:") && memswlimit
> 0 && memswusage
> 0) {
3060 snprintf(lbuf
, 100, "SwapFree: %8lu kB\n",
3061 (memswlimit
- memlimit
) - (memswusage
- memusage
));
3063 } else if (startswith(line
, "Slab:")) {
3064 snprintf(lbuf
, 100, "Slab: %8lu kB\n", 0UL);
3066 } else if (startswith(line
, "Buffers:")) {
3067 snprintf(lbuf
, 100, "Buffers: %8lu kB\n", 0UL);
3069 } else if (startswith(line
, "Cached:")) {
3070 snprintf(lbuf
, 100, "Cached: %8lu kB\n", cached
);
3072 } else if (startswith(line
, "SwapCached:")) {
3073 snprintf(lbuf
, 100, "SwapCached: %8lu kB\n", 0UL);
3078 l
= snprintf(cache
, cache_size
, "%s", printme
);
3080 perror("Error writing to cache");
3085 if (l
>= cache_size
) {
3086 fprintf(stderr
, "Internal error: truncated write to cache\n");
3097 d
->size
= total_len
;
3098 if (total_len
> size
) total_len
= size
;
3099 memcpy(buf
, d
->buf
, total_len
);
3108 free(memswlimit_str
);
3109 free(memswusage_str
);
3111 free(memswlimit_default_str
);
3112 free(memswusage_default_str
);
3117 * Read the cpuset.cpus for cg
3118 * Return the answer in a newly allocated string which must be freed
3120 static char *get_cpuset(const char *cg
)
3124 if (!cgfs_get_value("cpuset", cg
, "cpuset.cpus", &answer
))
3129 bool cpu_in_cpuset(int cpu
, const char *cpuset
);
3131 static bool cpuline_in_cpuset(const char *line
, const char *cpuset
)
3135 if (sscanf(line
, "processor : %d", &cpu
) != 1)
3137 return cpu_in_cpuset(cpu
, cpuset
);
3141 * check whether this is a '^processor" line in /proc/cpuinfo
3143 static bool is_processor_line(const char *line
)
3147 if (sscanf(line
, "processor : %d", &cpu
) == 1)
3152 static int proc_cpuinfo_read(char *buf
, size_t size
, off_t offset
,
3153 struct fuse_file_info
*fi
)
3155 struct fuse_context
*fc
= fuse_get_context();
3156 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3158 char *cpuset
= NULL
;
3160 size_t linelen
= 0, total_len
= 0, rv
= 0;
3161 bool am_printing
= false, firstline
= true, is_s390x
= false;
3162 int curcpu
= -1, cpu
;
3163 char *cache
= d
->buf
;
3164 size_t cache_size
= d
->buflen
;
3168 if (offset
> d
->size
)
3172 int left
= d
->size
- offset
;
3173 total_len
= left
> size
? size
: left
;
3174 memcpy(buf
, cache
+ offset
, total_len
);
3178 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3181 cg
= get_pid_cgroup(initpid
, "cpuset");
3183 return read_file("proc/cpuinfo", buf
, size
, d
);
3184 prune_init_slice(cg
);
3186 cpuset
= get_cpuset(cg
);
3190 f
= fopen("/proc/cpuinfo", "r");
3194 while (getline(&line
, &linelen
, f
) != -1) {
3198 if (strstr(line
, "IBM/S390") != NULL
) {
3204 if (strncmp(line
, "# processors:", 12) == 0)
3206 if (is_processor_line(line
)) {
3207 am_printing
= cpuline_in_cpuset(line
, cpuset
);
3210 l
= snprintf(cache
, cache_size
, "processor : %d\n", curcpu
);
3212 perror("Error writing to cache");
3216 if (l
>= cache_size
) {
3217 fprintf(stderr
, "Internal error: truncated write to cache\n");
3226 } else if (is_s390x
&& sscanf(line
, "processor %d:", &cpu
) == 1) {
3228 if (!cpu_in_cpuset(cpu
, cpuset
))
3231 p
= strchr(line
, ':');
3235 l
= snprintf(cache
, cache_size
, "processor %d:%s", curcpu
, p
);
3237 perror("Error writing to cache");
3241 if (l
>= cache_size
) {
3242 fprintf(stderr
, "Internal error: truncated write to cache\n");
3253 l
= snprintf(cache
, cache_size
, "%s", line
);
3255 perror("Error writing to cache");
3259 if (l
>= cache_size
) {
3260 fprintf(stderr
, "Internal error: truncated write to cache\n");
3271 char *origcache
= d
->buf
;
3274 d
->buf
= malloc(d
->buflen
);
3277 cache_size
= d
->buflen
;
3279 l
= snprintf(cache
, cache_size
, "vendor_id : IBM/S390\n");
3280 if (l
< 0 || l
>= cache_size
) {
3287 l
= snprintf(cache
, cache_size
, "# processors : %d\n", curcpu
+ 1);
3288 if (l
< 0 || l
>= cache_size
) {
3295 l
= snprintf(cache
, cache_size
, "%s", origcache
);
3297 if (l
< 0 || l
>= cache_size
)
3303 d
->size
= total_len
;
3304 if (total_len
> size
) total_len
= size
;
3306 /* read from off 0 */
3307 memcpy(buf
, d
->buf
, total_len
);
3318 static int proc_stat_read(char *buf
, size_t size
, off_t offset
,
3319 struct fuse_file_info
*fi
)
3321 struct fuse_context
*fc
= fuse_get_context();
3322 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3324 char *cpuset
= NULL
;
3326 size_t linelen
= 0, total_len
= 0, rv
= 0;
3327 int curcpu
= -1; /* cpu numbering starts at 0 */
3328 unsigned long user
= 0, nice
= 0, system
= 0, idle
= 0, iowait
= 0, irq
= 0, softirq
= 0, steal
= 0, guest
= 0;
3329 unsigned long user_sum
= 0, nice_sum
= 0, system_sum
= 0, idle_sum
= 0, iowait_sum
= 0,
3330 irq_sum
= 0, softirq_sum
= 0, steal_sum
= 0, guest_sum
= 0;
3331 #define CPUALL_MAX_SIZE BUF_RESERVE_SIZE
3332 char cpuall
[CPUALL_MAX_SIZE
];
3333 /* reserve for cpu all */
3334 char *cache
= d
->buf
+ CPUALL_MAX_SIZE
;
3335 size_t cache_size
= d
->buflen
- CPUALL_MAX_SIZE
;
3339 if (offset
> d
->size
)
3343 int left
= d
->size
- offset
;
3344 total_len
= left
> size
? size
: left
;
3345 memcpy(buf
, d
->buf
+ offset
, total_len
);
3349 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3352 cg
= get_pid_cgroup(initpid
, "cpuset");
3354 return read_file("/proc/stat", buf
, size
, d
);
3355 prune_init_slice(cg
);
3357 cpuset
= get_cpuset(cg
);
3361 f
= fopen("/proc/stat", "r");
3366 if (getline(&line
, &linelen
, f
) < 0) {
3367 fprintf(stderr
, "proc_stat_read read first line failed\n");
3371 while (getline(&line
, &linelen
, f
) != -1) {
3374 char cpu_char
[10]; /* That's a lot of cores */
3377 if (sscanf(line
, "cpu%9[^ ]", cpu_char
) != 1) {
3378 /* not a ^cpuN line containing a number N, just print it */
3379 l
= snprintf(cache
, cache_size
, "%s", line
);
3381 perror("Error writing to cache");
3385 if (l
>= cache_size
) {
3386 fprintf(stderr
, "Internal error: truncated write to cache\n");
3396 if (sscanf(cpu_char
, "%d", &cpu
) != 1)
3398 if (!cpu_in_cpuset(cpu
, cpuset
))
3402 c
= strchr(line
, ' ');
3405 l
= snprintf(cache
, cache_size
, "cpu%d%s", curcpu
, c
);
3407 perror("Error writing to cache");
3412 if (l
>= cache_size
) {
3413 fprintf(stderr
, "Internal error: truncated write to cache\n");
3422 if (sscanf(line
, "%*s %lu %lu %lu %lu %lu %lu %lu %lu %lu", &user
, &nice
, &system
, &idle
, &iowait
, &irq
,
3423 &softirq
, &steal
, &guest
) != 9)
3427 system_sum
+= system
;
3429 iowait_sum
+= iowait
;
3431 softirq_sum
+= softirq
;
3438 int cpuall_len
= snprintf(cpuall
, CPUALL_MAX_SIZE
, "%s %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3439 "cpu ", user_sum
, nice_sum
, system_sum
, idle_sum
, iowait_sum
, irq_sum
, softirq_sum
, steal_sum
, guest_sum
);
3440 if (cpuall_len
> 0 && cpuall_len
< CPUALL_MAX_SIZE
){
3441 memcpy(cache
, cpuall
, cpuall_len
);
3442 cache
+= cpuall_len
;
3444 /* shouldn't happen */
3445 fprintf(stderr
, "proc_stat_read copy cpuall failed, cpuall_len=%d\n", cpuall_len
);
3449 memmove(cache
, d
->buf
+ CPUALL_MAX_SIZE
, total_len
);
3450 total_len
+= cpuall_len
;
3452 d
->size
= total_len
;
3453 if (total_len
> size
) total_len
= size
;
3455 memcpy(buf
, d
->buf
, total_len
);
3467 static long int getreaperage(pid_t pid
)
3474 qpid
= lookup_initpid_in_store(pid
);
3478 ret
= snprintf(fnam
, 100, "/proc/%d", qpid
);
3479 if (ret
< 0 || ret
>= 100)
3482 if (lstat(fnam
, &sb
) < 0)
3485 return time(NULL
) - sb
.st_ctime
;
3488 static unsigned long get_reaper_busy(pid_t task
)
3490 pid_t initpid
= lookup_initpid_in_store(task
);
3491 char *cgroup
= NULL
, *usage_str
= NULL
;
3492 unsigned long usage
= 0;
3497 cgroup
= get_pid_cgroup(initpid
, "cpuacct");
3500 prune_init_slice(cgroup
);
3501 if (!cgfs_get_value("cpuacct", cgroup
, "cpuacct.usage", &usage_str
))
3503 usage
= strtoul(usage_str
, NULL
, 10);
3504 usage
/= 1000000000;
3515 char *name
, *cwd
= get_current_dir_name();
3521 len
= strlen(cwd
) + strlen("/iwashere") + 1;
3523 snprintf(name
, len
, "%s/iwashere", cwd
);
3525 fd
= creat(name
, 0755);
3532 * We read /proc/uptime and reuse its second field.
3533 * For the first field, we use the mtime for the reaper for
3534 * the calling pid as returned by getreaperage
3536 static int proc_uptime_read(char *buf
, size_t size
, off_t offset
,
3537 struct fuse_file_info
*fi
)
3539 struct fuse_context
*fc
= fuse_get_context();
3540 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3541 long int reaperage
= getreaperage(fc
->pid
);
3542 unsigned long int busytime
= get_reaper_busy(fc
->pid
), idletime
;
3543 char *cache
= d
->buf
;
3544 ssize_t total_len
= 0;
3551 if (offset
> d
->size
)
3555 int left
= d
->size
- offset
;
3556 total_len
= left
> size
? size
: left
;
3557 memcpy(buf
, cache
+ offset
, total_len
);
3561 idletime
= reaperage
- busytime
;
3562 if (idletime
> reaperage
)
3563 idletime
= reaperage
;
3565 total_len
= snprintf(d
->buf
, d
->size
, "%ld.0 %lu.0\n", reaperage
, idletime
);
3567 perror("Error writing to cache");
3571 d
->size
= (int)total_len
;
3574 if (total_len
> size
) total_len
= size
;
3576 memcpy(buf
, d
->buf
, total_len
);
3580 static int proc_diskstats_read(char *buf
, size_t size
, off_t offset
,
3581 struct fuse_file_info
*fi
)
3584 struct fuse_context
*fc
= fuse_get_context();
3585 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3587 char *io_serviced_str
= NULL
, *io_merged_str
= NULL
, *io_service_bytes_str
= NULL
,
3588 *io_wait_time_str
= NULL
, *io_service_time_str
= NULL
;
3589 unsigned long read
= 0, write
= 0;
3590 unsigned long read_merged
= 0, write_merged
= 0;
3591 unsigned long read_sectors
= 0, write_sectors
= 0;
3592 unsigned long read_ticks
= 0, write_ticks
= 0;
3593 unsigned long ios_pgr
= 0, tot_ticks
= 0, rq_ticks
= 0;
3594 unsigned long rd_svctm
= 0, wr_svctm
= 0, rd_wait
= 0, wr_wait
= 0;
3595 char *cache
= d
->buf
;
3596 size_t cache_size
= d
->buflen
;
3598 size_t linelen
= 0, total_len
= 0, rv
= 0;
3599 unsigned int major
= 0, minor
= 0;
3604 if (offset
> d
->size
)
3608 int left
= d
->size
- offset
;
3609 total_len
= left
> size
? size
: left
;
3610 memcpy(buf
, cache
+ offset
, total_len
);
3614 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3617 cg
= get_pid_cgroup(initpid
, "blkio");
3619 return read_file("/proc/diskstats", buf
, size
, d
);
3620 prune_init_slice(cg
);
3622 if (!cgfs_get_value("blkio", cg
, "blkio.io_serviced_recursive", &io_serviced_str
))
3624 if (!cgfs_get_value("blkio", cg
, "blkio.io_merged_recursive", &io_merged_str
))
3626 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_bytes_recursive", &io_service_bytes_str
))
3628 if (!cgfs_get_value("blkio", cg
, "blkio.io_wait_time_recursive", &io_wait_time_str
))
3630 if (!cgfs_get_value("blkio", cg
, "blkio.io_service_time_recursive", &io_service_time_str
))
3634 f
= fopen("/proc/diskstats", "r");
3638 while (getline(&line
, &linelen
, f
) != -1) {
3642 i
= sscanf(line
, "%u %u %71s", &major
, &minor
, dev_name
);
3646 get_blkio_io_value(io_serviced_str
, major
, minor
, "Read", &read
);
3647 get_blkio_io_value(io_serviced_str
, major
, minor
, "Write", &write
);
3648 get_blkio_io_value(io_merged_str
, major
, minor
, "Read", &read_merged
);
3649 get_blkio_io_value(io_merged_str
, major
, minor
, "Write", &write_merged
);
3650 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Read", &read_sectors
);
3651 read_sectors
= read_sectors
/512;
3652 get_blkio_io_value(io_service_bytes_str
, major
, minor
, "Write", &write_sectors
);
3653 write_sectors
= write_sectors
/512;
3655 get_blkio_io_value(io_service_time_str
, major
, minor
, "Read", &rd_svctm
);
3656 rd_svctm
= rd_svctm
/1000000;
3657 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Read", &rd_wait
);
3658 rd_wait
= rd_wait
/1000000;
3659 read_ticks
= rd_svctm
+ rd_wait
;
3661 get_blkio_io_value(io_service_time_str
, major
, minor
, "Write", &wr_svctm
);
3662 wr_svctm
= wr_svctm
/1000000;
3663 get_blkio_io_value(io_wait_time_str
, major
, minor
, "Write", &wr_wait
);
3664 wr_wait
= wr_wait
/1000000;
3665 write_ticks
= wr_svctm
+ wr_wait
;
3667 get_blkio_io_value(io_service_time_str
, major
, minor
, "Total", &tot_ticks
);
3668 tot_ticks
= tot_ticks
/1000000;
3670 memset(lbuf
, 0, 256);
3671 if (read
|| write
|| read_merged
|| write_merged
|| read_sectors
|| write_sectors
|| read_ticks
|| write_ticks
)
3672 snprintf(lbuf
, 256, "%u %u %s %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu %lu\n",
3673 major
, minor
, dev_name
, read
, read_merged
, read_sectors
, read_ticks
,
3674 write
, write_merged
, write_sectors
, write_ticks
, ios_pgr
, tot_ticks
, rq_ticks
);
3678 l
= snprintf(cache
, cache_size
, "%s", lbuf
);
3680 perror("Error writing to fuse buf");
3684 if (l
>= cache_size
) {
3685 fprintf(stderr
, "Internal error: truncated write to cache\n");
3695 d
->size
= total_len
;
3696 if (total_len
> size
) total_len
= size
;
3697 memcpy(buf
, d
->buf
, total_len
);
3705 free(io_serviced_str
);
3706 free(io_merged_str
);
3707 free(io_service_bytes_str
);
3708 free(io_wait_time_str
);
3709 free(io_service_time_str
);
3713 static int proc_swaps_read(char *buf
, size_t size
, off_t offset
,
3714 struct fuse_file_info
*fi
)
3716 struct fuse_context
*fc
= fuse_get_context();
3717 struct file_info
*d
= (struct file_info
*)fi
->fh
;
3719 char *memswlimit_str
= NULL
, *memlimit_str
= NULL
, *memusage_str
= NULL
, *memswusage_str
= NULL
,
3720 *memswlimit_default_str
= NULL
, *memswusage_default_str
= NULL
;
3721 unsigned long memswlimit
= 0, memlimit
= 0, memusage
= 0, memswusage
= 0, swap_total
= 0, swap_free
= 0;
3722 ssize_t total_len
= 0, rv
= 0;
3724 char *cache
= d
->buf
;
3727 if (offset
> d
->size
)
3731 int left
= d
->size
- offset
;
3732 total_len
= left
> size
? size
: left
;
3733 memcpy(buf
, cache
+ offset
, total_len
);
3737 pid_t initpid
= lookup_initpid_in_store(fc
->pid
);
3740 cg
= get_pid_cgroup(initpid
, "memory");
3742 return read_file("/proc/swaps", buf
, size
, d
);
3743 prune_init_slice(cg
);
3745 if (!cgfs_get_value("memory", cg
, "memory.limit_in_bytes", &memlimit_str
))
3748 if (!cgfs_get_value("memory", cg
, "memory.usage_in_bytes", &memusage_str
))
3751 memlimit
= strtoul(memlimit_str
, NULL
, 10);
3752 memusage
= strtoul(memusage_str
, NULL
, 10);
3754 if (cgfs_get_value("memory", cg
, "memory.memsw.usage_in_bytes", &memswusage_str
) &&
3755 cgfs_get_value("memory", cg
, "memory.memsw.limit_in_bytes", &memswlimit_str
)) {
3757 /* If swap accounting is turned on, then default value is assumed to be that of cgroup / */
3758 if (!cgfs_get_value("memory", "/", "memory.memsw.limit_in_bytes", &memswlimit_default_str
))
3760 if (!cgfs_get_value("memory", "/", "memory.memsw.usage_in_bytes", &memswusage_default_str
))
3763 memswlimit
= strtoul(memswlimit_str
, NULL
, 10);
3764 memswusage
= strtoul(memswusage_str
, NULL
, 10);
3766 if (!strcmp(memswlimit_str
, memswlimit_default_str
))
3768 if (!strcmp(memswusage_str
, memswusage_default_str
))
3771 swap_total
= (memswlimit
- memlimit
) / 1024;
3772 swap_free
= (memswusage
- memusage
) / 1024;
3775 total_len
= snprintf(d
->buf
, d
->size
, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
3777 /* When no mem + swap limit is specified or swapaccount=0*/
3781 FILE *f
= fopen("/proc/meminfo", "r");
3786 while (getline(&line
, &linelen
, f
) != -1) {
3787 if (startswith(line
, "SwapTotal:")) {
3788 sscanf(line
, "SwapTotal: %8lu kB", &swap_total
);
3789 } else if (startswith(line
, "SwapFree:")) {
3790 sscanf(line
, "SwapFree: %8lu kB", &swap_free
);
3798 if (swap_total
> 0) {
3799 l
= snprintf(d
->buf
+ total_len
, d
->size
- total_len
,
3800 "none%*svirtual\t\t%lu\t%lu\t0\n", 36, " ",
3801 swap_total
, swap_free
);
3805 if (total_len
< 0 || l
< 0) {
3806 perror("Error writing to cache");
3812 d
->size
= (int)total_len
;
3814 if (total_len
> size
) total_len
= size
;
3815 memcpy(buf
, d
->buf
, total_len
);
3820 free(memswlimit_str
);
3823 free(memswusage_str
);
3824 free(memswusage_default_str
);
3825 free(memswlimit_default_str
);
3829 static off_t
get_procfile_size(const char *which
)
3831 FILE *f
= fopen(which
, "r");
3834 ssize_t sz
, answer
= 0;
3838 while ((sz
= getline(&line
, &len
, f
)) != -1)
3846 int proc_getattr(const char *path
, struct stat
*sb
)
3848 struct timespec now
;
3850 memset(sb
, 0, sizeof(struct stat
));
3851 if (clock_gettime(CLOCK_REALTIME
, &now
) < 0)
3853 sb
->st_uid
= sb
->st_gid
= 0;
3854 sb
->st_atim
= sb
->st_mtim
= sb
->st_ctim
= now
;
3855 if (strcmp(path
, "/proc") == 0) {
3856 sb
->st_mode
= S_IFDIR
| 00555;
3860 if (strcmp(path
, "/proc/meminfo") == 0 ||
3861 strcmp(path
, "/proc/cpuinfo") == 0 ||
3862 strcmp(path
, "/proc/uptime") == 0 ||
3863 strcmp(path
, "/proc/stat") == 0 ||
3864 strcmp(path
, "/proc/diskstats") == 0 ||
3865 strcmp(path
, "/proc/swaps") == 0) {
3867 sb
->st_mode
= S_IFREG
| 00444;
3875 int proc_readdir(const char *path
, void *buf
, fuse_fill_dir_t filler
, off_t offset
,
3876 struct fuse_file_info
*fi
)
3878 if (filler(buf
, "cpuinfo", NULL
, 0) != 0 ||
3879 filler(buf
, "meminfo", NULL
, 0) != 0 ||
3880 filler(buf
, "stat", NULL
, 0) != 0 ||
3881 filler(buf
, "uptime", NULL
, 0) != 0 ||
3882 filler(buf
, "diskstats", NULL
, 0) != 0 ||
3883 filler(buf
, "swaps", NULL
, 0) != 0)
3888 int proc_open(const char *path
, struct fuse_file_info
*fi
)
3891 struct file_info
*info
;
3893 if (strcmp(path
, "/proc/meminfo") == 0)
3894 type
= LXC_TYPE_PROC_MEMINFO
;
3895 else if (strcmp(path
, "/proc/cpuinfo") == 0)
3896 type
= LXC_TYPE_PROC_CPUINFO
;
3897 else if (strcmp(path
, "/proc/uptime") == 0)
3898 type
= LXC_TYPE_PROC_UPTIME
;
3899 else if (strcmp(path
, "/proc/stat") == 0)
3900 type
= LXC_TYPE_PROC_STAT
;
3901 else if (strcmp(path
, "/proc/diskstats") == 0)
3902 type
= LXC_TYPE_PROC_DISKSTATS
;
3903 else if (strcmp(path
, "/proc/swaps") == 0)
3904 type
= LXC_TYPE_PROC_SWAPS
;
3908 info
= malloc(sizeof(*info
));
3912 memset(info
, 0, sizeof(*info
));
3915 info
->buflen
= get_procfile_size(path
) + BUF_RESERVE_SIZE
;
3917 info
->buf
= malloc(info
->buflen
);
3918 } while (!info
->buf
);
3919 memset(info
->buf
, 0, info
->buflen
);
3920 /* set actual size to buffer size */
3921 info
->size
= info
->buflen
;
3923 fi
->fh
= (unsigned long)info
;
3927 int proc_access(const char *path
, int mask
)
3929 /* these are all read-only */
3930 if ((mask
& ~R_OK
) != 0)
3935 int proc_release(const char *path
, struct fuse_file_info
*fi
)
3937 do_release_file_info(fi
);
3941 int proc_read(const char *path
, char *buf
, size_t size
, off_t offset
,
3942 struct fuse_file_info
*fi
)
3944 struct file_info
*f
= (struct file_info
*) fi
->fh
;
3947 case LXC_TYPE_PROC_MEMINFO
:
3948 return proc_meminfo_read(buf
, size
, offset
, fi
);
3949 case LXC_TYPE_PROC_CPUINFO
:
3950 return proc_cpuinfo_read(buf
, size
, offset
, fi
);
3951 case LXC_TYPE_PROC_UPTIME
:
3952 return proc_uptime_read(buf
, size
, offset
, fi
);
3953 case LXC_TYPE_PROC_STAT
:
3954 return proc_stat_read(buf
, size
, offset
, fi
);
3955 case LXC_TYPE_PROC_DISKSTATS
:
3956 return proc_diskstats_read(buf
, size
, offset
, fi
);
3957 case LXC_TYPE_PROC_SWAPS
:
3958 return proc_swaps_read(buf
, size
, offset
, fi
);
3964 static void __attribute__((constructor
)) collect_subsystems(void)
3970 if ((f
= fopen("/proc/self/cgroup", "r")) == NULL
) {
3971 fprintf(stderr
, "Error opening /proc/self/cgroup: %s\n", strerror(errno
));
3974 while (getline(&line
, &len
, f
) != -1) {
3977 p
= strchr(line
, ':');
3982 p2
= strrchr(p
, ':');
3987 /* With cgroupv2 /proc/self/cgroup can contain entries of the
3988 * form: 0::/ This will cause lxcfs to fail the cgroup mounts
3989 * because it parses out the empty string "" and later on passes
3990 * it to mount(). Let's skip such entries.
3995 if (!store_hierarchy(line
, p
))
4006 static void __attribute__((destructor
)) free_subsystems(void)
4010 for (i
= 0; i
< num_hierarchies
; i
++)
4012 free(hierarchies
[i
]);